Spatiotemporal dynamics and environmental drivers of older pedestrian traffic crashes in Hong Kong: a spatial machine learning approach

Older pedestrians are increasingly vulnerable to traffic crashes, particularly in high-density and aging urban environments. Despite growing academic attention, few studies have systematically examined the long-term spatiotemporal dynamics and complex environmental impacts of older pedestrian traffic crashes (OPTC) at fine spatial scales. This study addresses these gaps by analyzing a decade of crash data (2014–2023) in Hong Kong Special Administrative Region, using the street block as the analytical unit. Multi-source datasets, including police-reported crash records, census data, points of interest, and Google Street View imagery, were integrated into a comprehensive analytical framework that combines spatiotemporal hotspot detection, geographically weighted random forest (GWRF), and SHapley Additive exPlanations (SHAP) for model interpretation. The results reveal persistent and intensifying crash hotspots in core urban districts, as well as significant upward trends in suburban and newly developed areas such as Tai Po, Tuen Mun, and Yuen Long. Key variables, including the density of shopping and healthcare facilities, street greenery, and visual complexity, exhibit spatially heterogeneous, nonlinear, and threshold-based effects. Built environment variables show increased crash risks when critical density thresholds are exceeded, while the protective effects of street environment features become significant only beyond baseline levels. Moreover, interaction analysis further uncovers both risk-amplifying and risk-mitigating synergies between built and street environment attributes. Overall, these findings provide spatially refined and interpretable evidence on the mechanisms underlying OPTC and offer valuable insights for targeted, age-friendly interventions in urban planning and transport policy.